The Impact of Clock Jitter on MIDI syncable DAWs - Technical Report E-RM - Erfindungsbüro Rest & Maier
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Impact of Clock Jitter on MIDI
syncable DAWs
Technical Report
Maximilian Rest
E-RM - Erfindungsbüro Rest & Maier
info@midiclock.de
www.midiclock.de
Berlin, Februrary 2014
(updated 4.3.2014)Contents
1 Introduction 1
2 Problem Setting 2
3 Verify Synchronisation Problems between DAWs 3
4 Enhance Synchronisation with the E-RM midiclock 7
5 Conclusion 11
Appendix 12
References 16
1 Introduction
“How sour sweet music is, when time is broke and no proportion
kept!”
– William Shakespeare, King Richard II, 1595
Already more than 400 years ago, Shakespeare knew about the effects of jitter
and phasing in musical performances, and it tells in a nutshell what contemporary
musicians still have to take care of today - namely the tight synchronisation of
different tracks when using multiple computers as digital audio workstations.
The report at hand explaines a setup of synchronising multiple DAWs, analyses
problems most musicians face with MIDI Clock and shows how the E-RM midiclock
can greatly improve overall timing.
Although Live from Ableton AG has been used for the experiments in this paper,
all findings hold true in general for all other Music Production Softwares.
If you are interested in the experimental results only, you may jump to Section 5 on
Page 11.
1The Impact of Clock Jitter on MIDI syncable DAWs
2 Problem Setting
The sequencer under examination will be Live from Ableton AG, one of the most
popular music production softwares currently on the market [4],[8],[12].
Like musicians that play classical instruments, users of DAWs also want to play
together with other artists. This is why Live has capabilities to synchronise software
instances on different computers. One of the easiest solutions is to use the Musical
Instrument Digital Interface (MIDI). The MIDI specification offers the use of MIDI
Clock signals that keep all connected devices in sync like the drummer in a band
[13, p. 30].
MIDI is technically an unidirectional serial interface that always needs a master who
sends data, and a slave, who receives it. Live can either be configured as a master or
a slave concerning MIDI Clock data. A common way to synchronise multiple DAWs
is depicted in Figure 1.
“Specifically, the master sends 24 MIDI Clocks, spaced at equal intervals, during
every quarter note interval.” [1]. This enables the slave to extract tempo and phase
information from the incoming Beat Clock signals and play back all audio tracks
synchronised to the master.
Figure 1: Usual MIDI Beat Clock distribution between multiple DAWs
MIDI Syncronisation and Audio Jitter
A problem which a lot of artists confront when trying to merge audio tracks played
by different DAWs (e.g. as in a setup as shown in Fig. 1) is that the resulting
synchronisation lacks accuracy.
Although the concept of MIDI Clock appeals through theoretical simplicity, practi-
cally the audio tracks from the slave DAW often loose synchronisation and are not
played tightly to the master DAW. Ableton introduced clock slave improvements in
Live 8.4.1b1 back in 2011, but states:
Erfindungsbüro Rest & Maier 2The Impact of Clock Jitter on MIDI syncable DAWs
“The stability of the slave algorithm still depends on your system and
the quality of your master clock. If you have a real bad master clock or
bad MIDI drivers, you probably will not see much of an improvement.”
[11]
3 Verify Synchronisation Problems between DAWs
To verify this phaenomenon on a scientifical level, a test setup as shown in Fig-
ure 2 is evaluated. Two DAWs with Live 9 are connected through their respec-
tive MIDI ports. The MIDI master clock is converted to an audio signal [5] and
recorded together with the audio output of the slave DAW with a sampling fre-
quency fs = 96kHz.
Figure 2: Test setup to verify synchronisation problems between DAWs
Both master and slave DAW are put under 50% CPU load to ensure real-life cir-
cumstances. The master entity of Live is set to 120 Beats per Minute (BPM) and
streams MIDI Clock Ticks to it’s MIDI OUT port. On the slave DAW - which locks
its playback to the incoming MIDI Clock stream - a short click-sound is played eight
times per quarter note.
Two tests with MIDI slave receivers in completely different price ranges are per-
formed to compare the gathered data. In Test A, a professional MIDI receiver with
a dedicated software driver is used, in Test B a simple MIDI to USB converter with
the operating systems class compliant driver is installed. The detailed device config-
urations are shown in Table 3, page 15. The test duration is 5 minutes and the audio
files are recorded in an uncompressed, sample accurate format for further processing.
Erfindungsbüro Rest & Maier 3The Impact of Clock Jitter on MIDI syncable DAWs
Test Results
The data has been numerically analyzed with the free software package GNU Octave.
Figure 3 and 4 show BPM(t) behaviour for Test A and Test B. Tempo drops
and bursts of approx. ±0.5BPM centered around the desired playback rate of
BPMset = 120 can be observed in both graphs, but more frequent in Test B where
a MIDI Interface without a dedicated driver is used.
120.6
Tempo rate [BPM]
120.4
120.2
120
119.8
119.6
119.4
119.2
0 50 100 150 200 250
Time [s]
Figure 3: BPM over Time for Test A
120.6
Tempo rate [BPM]
120.4
120.2
120
119.8
119.6
119.4
119.2
0 50 100 150 200 250
Time [s]
Figure 4: BPM over Time for Test B
Moreover, also outside the range of the bursts the tempo constantly oscillates in
3
both curves with ± 100 BPM. Statistical data can be found in Table 1.
A look at Figures 5 and 6 shows at a glance what artists complain about: The notes
of the slave DAW are not played on their desired playback position, which causes
the perception of inaccurate synchronisation.
Erfindungsbüro Rest & Maier 4The Impact of Clock Jitter on MIDI syncable DAWs
10
8
Beat Offset [ms]
6
4
2
0
-2
0 50 100 150 200 250
Time [s]
Figure 5: Beat Offset over Time for Test A
2
Beat Offset [ms]
0
-2
-4
-6
-8
0 50 100 150 200 250
Time [s]
Figure 6: Beat Offset over Time for Test B
The problem is worse with the cheap MIDI interface than with the professional grade
device, but both setups have severe timing issues. Whereas the graph from Test A
(Fig. 5) at least settles back to a constant offset of roughly −1ms after drops and
bursts in the tempo rate, the graph from Test B (Fig. 6) settles at −7ms, −7.5ms
and −8ms.
Setup B most probably even looses MIDI Ticks, the equation to calculate Beat
Offset acts like an integrator and the constant negative offest increases over time
(see appendix, p. 13). The PLL unlocks on a lost MIDI Tick and locks again
afterwards, in an unknown phase state, which results in a different offset [2, p. 103].
The histograms in Figure 7 show the distribution of Beat Offset for both tests. The
interpretation of the mean value deserves our particular interest, as the integration
constant depends on the start position of the measured data and is not set at the
beginning of the data processing.
Test A may look okay at the first glance due to its high peak at a constant offset of
−1ms, but nevertheless the standard deviation is 2.28ms and peaks of up to 9ms
occur (Tab. 1).
For Test B, the spread of the min and max values is almost the same as in Test A,
and so is the standard deviation. Differences can be seen in the histrogram, which
shows a more widespread distribution with a much lower peak at −8ms relative to
the other Beat Offsets which occupy a larger area.
The Ableton Reference Manual states that “(Beat) Jitter, much more so than la-
tency, creates the feeling that (..) timing is “sloppy“ or “loose.“ ” [3, p. 592]. The
Erfindungsbüro Rest & Maier 5The Impact of Clock Jitter on MIDI syncable DAWs
Setups in Test A and B would most probably fail a listening test, as a Beat Offset
of −8ms equals 13% of the ideal period of Tideal /fs = 62.5ms between adjacent notes.
The cycle-to-cycle jitter of the MIDI Clock signal from the master DAW can be
seen in Figure 8. It is a direct measure of instantaneous jumps of frequency in the
clock stream [7, p. 3]. MIDI data from Test A and Test B looked almost the same
regarding cycle-to-cycle jitter so that only data from Test A is shown.
The ideal period between adjacent MIDI Clock Ticks at BPMset is 60s/(120 · 24) =
20.1ms, cycle-to-cycle jitter has peaks of −38ms (Tab. 1).
In this magnitude, C2C jitter obviously exceeds the lock-range of the slave PLL,
which gets unlocked from the master tempo and cycles start to slip through [10, p.
PLL/6] [2, p. 89]. Unfortunately, one can’t precisely distinguish between lock-outs
due to ticks that got lost in the slave DAW and those due to excessive cycle-to-cycle
jitter.
Test A Test B
3000 1000
2500 800
Number of Beats
Number of Beats
2000
600
1500
400
1000
500 200
0 0
-8 -6 -4 -2 0 2 4 6 8 -8 -6 -4 -2 0 2 4 6 8
Beat Offset [ms] Beat Offset [ms]
Figure 7: Histograms of Beat Offset distribution for Test A and Test B
DAW Masterclock Signal
1400
1200
Number of Ticks
1000
800
600
400
200
0
-30 -20 -10 0 10 20 30
Cycle to Cycle Jitter [ms]
Figure 8: Histogram of DAW generated MIDI Master Clock cycle-to-cycle jitter
Erfindungsbüro Rest & Maier 6The Impact of Clock Jitter on MIDI syncable DAWs
√
Test: Data N unit → mean var min max
A: Tempo Rate 4500 BPM 120.000 0.114 119.26 120.62
A: Beat Offset 4500 ms 0.231 2.278 -1.51 8.99
B: Tempo Rate 4500 BPM 120.000 0.179 119.23 120.68
B: Beat Offset 4500 ms -5.066 2.869 -8.51 2.36
A+B: MIDI C2C 13500 ms 0 8.433 -38.36 22.24
Table 1: Statistical data for Test A and Test B
4 Enhance Synchronisation with the E-RM midiclock
To show the improvements of DAW-Sync by the E-RM midiclock, a test setup as
shown in Figure 9 is used. The E-RM midiclock acts as the Master Clock, just like
the Master DAW in Section 3, Fig. 2.
Figure 9: Test setup to verify synchronisation enhancement between DAWs by the
E-RM midiclock
Again, two different MIDI slave receivers are used, the professional grade receiver
for Test C, the cheap & simple one for Test D. The rest of the devices is also the
same as in in Section 3 (see Table 3 on page 15 for details).
The recording frequency is fs = 96kHz, the test duration 5 minutes to make the
results comparable to Test A and Test B.
As the sampling frequency of the audio recorder in the setup of Fig. 9 is only fs =
96kHz, the MIDI Clock stream from the E-RM midiclock is analyzed furthermore
Erfindungsbüro Rest & Maier 7The Impact of Clock Jitter on MIDI syncable DAWs
with a logic analyzer at fs = 24M Hz, as the maximum time resolution at fs =
96kHz is only Tres = 1/fs = 10.42us.
Test Results
BPM(t) is calculated as in Equation 6 (Appendix, p. 13) and compared against the
former results from Test A and Test B without a stable MIDI Clock signal.
Most obviously, no drops and bursts in the tempo rate can be found in the data
for both Test C/D (Fig. 10 and 11). The tempo rate is stable at 120BPM ± 0.05%
(Table 2).
120.6 Test A
Test C
Tempo rate [BPM]
120.4
120.2
120
119.8
119.6
119.4
119.2
0 50 100 150 200 250
Time [s]
Figure 10: BPM over Time for Test C
120.6 Test B
Test D
Tempo rate [BPM]
120.4
120.2
120
119.8
119.6
119.4
119.2
0 50 100 150 200 250
Time [s]
Figure 11: BPM over Time for Test D
There is a serious Beat Offset enhancement as can be seen from Figure 12 and 13.
No more dropouts, no loosing-track of the desired beat frequency. The stable MIDI
Clock signal drastically enhances synchronisation of the slaved DAW.
Erfindungsbüro Rest & Maier 8The Impact of Clock Jitter on MIDI syncable DAWs
Test A
8 Test C
Beat Offset [ms]
6
4
2
0
-2
0 50 100 150 200 250
Time [s]
Figure 12: Beat Offset over Time for Test C
2 Test B
Test D
Beat Offset [ms]
0
-2
-4
-6
-8
0 50 100 150 200 250
Time [s]
Figure 13: Beat Offset over Time for Test D
The gathered MIDI data reveals MIDI cycle-to-cycle jitter, the histogram in Figure
14 shows the result.
Enhanced Masterclock Signal
1.5k
Number of Ticks
1k
0.5k
0
-0.4 -0.2 0 0.2 0.4
Cycle to Cycle Jitter [us]
Figure 14: Histogram of midiclock cycle-to-cycle jitter in [us]
The former maximum MIDI Clock jitter of 38ms is reduced by three magnitudes
down to 400ns by the E-RM midiclock.
A look at the histogram plots of Beat Offset in Figure 15 reveals even more evidence
that a stable MIDI Clock signal improves audio playback on a slaved DAW. The
distributions are almost gaussian and centered very narrow around their mean value.
Erfindungsbüro Rest & Maier 9The Impact of Clock Jitter on MIDI syncable DAWs
A relative enhancement ratio is defined by comparing the highest magnitudes of
Beat Offset of related tests:
max(meanold − minold , maxold − meanold )
e= (1)
max(meannew − minnew , minnew − meanmax )
For Test A/C, Beat Offset enhances by a factor of e = 79.37, for Test B/D by
e = 29.94.
As the Clock data from the E-RM midiclock which gets fed into the slave DAW
can be considered almost ’jitter-free’, the biggest portion of remaining beat jitter is
inducted by the slave DAW and it’s MIDI interface and driver.
It is important to note that the relative enhancement ratio e is higher for Test C.
A professional MIDI Interface with dedicated drivers apparently adds less jitter to
the incoming messages.
Test C Test D
500 500
400 400
Number of Beats
Number of Beats
300 300
200 200
100 100
0 0
-150 -100 -50 0 50 100 150 -200 -100 0 100 200
Beat Offset [us] Beat Offset [us]
Figure 15: Normalized Histograms of Audio TIE distribution for Test C and Test D
√
Test: Data N unit → mean var min max
C: Tempo Rate 4500 BPM 120.020 0.018 119.96 120.06
C: Beat Offset 4500 us -33.895 37.841 -144.25 71.22
D: Tempo Rate 4500 BPM 120.020 0.018 119.96 120.06
D: Beat Offset 4500 us -5.374 93.755 -253.44 228.49
Table 2: Statistical data for Test C and Test D
Erfindungsbüro Rest & Maier 10The Impact of Clock Jitter on MIDI syncable DAWs 5 Conclusion To ensure tight synchronisation of MIDI syncable DAWs, a stable master clock signal must be provided. With the E-RM midiclock, a simple solution is available, which eliminates frequently changing beat phasing and forces all slaved DAWs to follow the given tempo. Reducing the former maximum MIDI Clock jitter of 38ms by three magnitudes down to 400ns caused a cut down of Beat Offset spread from 10ms by a factor of 48 to 210us with a professional MIDI receiver. With the E-RM midiclock, an endless number of slaves can be connected using MIDI distribution boxes, which will all run in sync with the master clock and thus in sync with each other. Erfindungsbüro Rest & Maier 11
The Impact of Clock Jitter on MIDI syncable DAWs
Appendix
Introduction to Jitter
The problem which one confronts when audio playback on a slaved DAW is out of
sync is called jitter:
“Jitter is defined as the (..) deviation of a signal’s transition time from
its ideal position in time.” [6, p. 122]
Figure 16: Definition of jitter [6, p. 128]
In the case of audio synchronisation, the N notes recorded from the slave DAW can
be treated as the signal under examination. They are compared to an imaginary ref-
erence clock with ideal edge positions, predefined by the notes that should regularily
be played at the given tempo [6, pp 124-125,127].
As defined by Maichen in Digital Timing Measurements [6], this jitter is generally
called “Time Interval Error” (TIE, see Fig. 16). In this case, TIE is the deviation
of the actual playback time of a note from its ideal position.
Figure 17: Different jitter trend plots [taken from 6, p. 128]
Erfindungsbüro Rest & Maier 12The Impact of Clock Jitter on MIDI syncable DAWs
In the evaluation it became evident that the gathered MIDI data can’t be used as a
reference clock for audio jitter calculation because of it’s own jitter. Since there is
no other reference clock recorded in the setup, time interval error must be calculated
round the back by period jitter and the desired period between two notes.
For that purpose all positions pos(n) of the notes n ∈ [1 : N ] are extracted with nu-
merical software from the respective recording. As the audio files are uncompressed
and sample accurate, the returned position pos(n) is a sample number in the file.
To obtain the period T (n) (in samples) between two notes n and n + 1 from the
position data, adjacent positions must be subtracted:
T (n) = pos(n + 1) − pos(n) , n ∈ [1 : N − 1] (2)
At the playback tempo of BP Mset = 120 in the tests, the desired period Tideal
between two adjacent notes is:
fs 96000 · 60s
Tideal = = = 6000 (3)
8 · BP Mset 8 · 120 · s
The factor 8 in the denominator represents the eight notes per quarter with BPM
in general defined as the rate of quarter notes per minute.
Period jitter ∆T (n) is then calculated as
∆T (n) = T (n) − Tideal , n ∈ [1 : N − 1] (4)
Like shown in Figure 17, time intervall error ∆pos (n) can now easily be calculated
from ∆T (n) by summation [9, p. 14].
n
X
∆pos (n) = ∆T (i) , n ∈ [1 : N − 1] (5)
i=1
Tideal probably needs some adjustment by a few per mill to take the actual playback
tempo into account (for example 120.01 instead of 120.00 BPM). Otherwise, the
summation in Equ. 5 will add up this error and run high or low.
T (n) and ∆pos (n) are now a direct measure for the actual tempo rate of the slave and
the offset of the desired beat position. To make the results easily comprehensible,
T (n) and ∆pos (n) are translated into BPM and a Beat Offset in seconds respectively.
BPM(t) as the actually measured playback rate at the time t is obtained by rear-
ranging Equ. 3
fs · 60s pos(n)
BPM(t) = with t= , n ∈ [1 : N − 1] (6)
8 · T (n) fs
Beat Offset in seconds is gained by converting the sample count to time:
∆pos (n) pos(n)
TIE(t) = also with t= , n ∈ [1 : N − 1] (7)
fs fs
Erfindungsbüro Rest & Maier 13The Impact of Clock Jitter on MIDI syncable DAWs
In addition to the representation on the timeline, histograms of both BPM(t) and
TIE(t) are plotted, as this gives a clue on the distribution of jitter [6, pp 129,132].
The MIDI Clock Ticks that are fed into the slave DAW are analyzed seperatly.
As Live is synchronised to incoming clock pulses that travel through a complex
computer system, some sort of clock smoothing and phase correction algorithm is
used to calculate playback position and rate [11].
Most probably, a Software Phase-Locked-Loop (SPLL) accomplishes the desired
filtering. As stated in Digital Timing Measurements, Section 8.2.1.4, a very useful
parameter to estimate the stability of PLLs is cycle-to-cycle jitter (C2C). It is a
direct measure of instantaneous jumps of frequency in the clock stream [7, p. 3].
Large deviations from the ideal clock tick position result in a low frequency jitter
which can not be filtered out by the low-pass loop filter anymore, the PLL gets
unlocked [10, p. PLL/6].
MIDI Clock Tick cycle-to-cycle jitter C2C(n) can be calculated from the NM posi-
tions pos(n) of the MIDI Ticks
C2C(n) = [pos(n + 2) − pos(n + 1)] − [pos(n + 1) − pos(n)] (8)
with n ∈ [1 : NM − 2] [9, p. 14].
Erfindungsbüro Rest & Maier 14The Impact of Clock Jitter on MIDI syncable DAWs
Devices used during the Test
Function Machine Configuration MIDI Interface Soundcard
Test A
Master Samsung Q35 ESI
CD 1.66Ghz, 1GB Ram MIDIMATE II
Windows XP, Live 9.0.2 USB 2.0
Slave Apple MacBook Pro RME Fireface RME
i7 1.66Ghz, 4GB Ram UCX, FW400 Fireface UCX
Mac OS X, Live 9.1.1 (dedicated driver)
Test B
Master like Test A
Slave Elektron
like Test A TM-1, USB 2.0 like Test A
(class compliant)
Test C
Master E-RM midiclock
Slave like Test A
Logic Saleae Logic
Analyzer 8ch, max. 24MHz sampling
Test D
Master E-RM midiclock
Slave like Test B
L.A. like Test B
All Tests
Recorder IBM X61
C2D 1.6Ghz, 4GB Ram onboard
Linux Mint, Audacity 2.0.0
Table 3: Devices used to verify Synchronisation problems between DAWs
Erfindungsbüro Rest & Maier 15The Impact of Clock Jitter on MIDI syncable DAWs
References
[1] Anonymous. Syncing Sequence Playback.
http://home.roadrunner.com/~jgglatt/tech/midispec/seq.htm; 2014.
[2] Best, Roland E. Phase-locked Loops - Design, Simulation and Applications.
6th Edition. McGraw-Hill, 2007. isbn: 978-0-07-1-49375-8.
[3] Dennis DeSantis et al. Ableton Reference Manual Version 9. 2013.
[4] Digital Music Doctor LLC. Music Software Internet Popularity.
http://www.digitalmusicdoctor.com/popularity; 2013.
[5] Innerclock Systems Pty Ltd. MIDI Clock to Audio circuitry.
www.innerclocksystems.com/NewICSClockWatch.html; 2014.
[6] Wolfgang Maichen. Digital Timing Measurements - From Scopes and Probes
to Timing and Jitter. Frontiers in Electronic Testing, Vol. 33. Springer, 2006.
isbn: 978-0-387-31418-1.
[7] Howell Mitchell. A Primer on Jitter, Jitter Measurement and Phase Locked
Loops. Application Note 687 - Rev. 0.1 5/12. Silicon Labs, 2012.
[8] MusicRadar. The 15 best DAW software apps in the world today.
http : / / www . musicradar . com / tuition / tech / the - 16 - best - daw -
software-apps-in-the-world-today-238905/15; 2012.
[9] PCI-SIG. PCI ExpressTM Jitter Modeling. White Paper - Revision 1.0RD.
2004.
[10] Prof. Dr.-Ing Klaus Petermann and Dr.-Ing Christian-Alexander Bunge. Hochfre-
quenztechnik II - Vorlesungsskript. 2012.
[11] Nico Starke. MIDI clock (slave) improvements in Live 8.2.4b1.
https://forum.ableton.com/viewtopic.php?f=1&t=165235; 2011.
[12] TechMedia Network. 2014 Audio Production Software Product Comparisons.
http://audio-production-software-review.toptenreviews.com; 2014.
[13] The MIDI Manufacturers Association. MIDI 1.0 Detailed Specification. Doc-
ument Version 4.2. 1995.
Erfindungsbüro Rest & Maier 16You can also read
